1
Occupational Licensing in the European Union:
Coverage and Wage Effects1
Maria Koumenta2, Mario Pagliero3
December 2017
We present the first EU-wide study on the prevalence and labour market impact of occupational
regulation in the EU. Drawing on a new EU Survey of Regulated Occupations, we find that
licensing affects about 22 percent of workers in the EU, although there is significant variability
across member states and occupations. On average, licensing is associated with a 4 percent
higher hourly wages. Using decomposition techniques we show that rent capture accounts for
one third of this effect and the remaining is attributed to signalling. We find considerable
heterogeneity in the wage gains by occupation and level of educational attainment. Finally,
occupational licensing increases wage inequality. After accounting for composition effects,
licensing increases the standard deviation of wages by about 0.02 log points.
JEL codes: J44, J31,
Keywords: Occupational regulation, licensing, wages.
1 We would like to thank Mindy Marks, Mark Law, Robert Thornton and seminar audiences at the European
Commission, 2016 AEA Meetings, 2017 Occupational Regulation conferences for helpful comments and
discussions. The opinions expressed in this paper and all remaining errors are those of the authors alone. 2 Queen Mary, University of London. 3 University of Turin, Collegio Carlo Alberto, and CEPR.
2
1. Introduction
Occupational regulation is a labour market institution that has attracted considerable debate
within academic and policy cycles. The current policy interest derives from its potential to
serve as a strong incentive for employers and workers to invest more heavily in skills and as a
means to safeguard consumer interests. However, economists have also pointed out the
potential role of rent-seeking activities by occupational groups and have warned against its
labour market effects (e.g. Friedman 1962; Maurizi 1974; Graddy 1991; Shapiro 1986). While
there is now a well-developed empirical literature in the US assessing the labour market
outcomes of occupational regulation, the paucity of such evidence in the European context is
striking. This is a surprising omission given that, as we shall show below, entry to a significant
proportion of EU jobs is restricted through licensing. As such, the importance of this labour
market institution extends beyond academic curiosity and deserves more attention than it
currently receives.
On the other hand, this gap in the literature is not entirely unjustifiable. Since recently in Europe
researchers have been lacking comprehensive data that identifies regulated individuals, the
characteristics of the regulatory regime they are subject to and documents their individual and
labour market characteristics. This paper addresses this gap. We explore the first ever EU
Survey of Regulated Occupations (EU-SOR), which consists of a nationally representative
sample that covers the labor force within the EU28 member states and asks detailed questions
about occupational regulation. We are interested in three key themes: incidence of regulation
in the EU labour market, its effect on wage determination and its effect on wage inequality.
Prior to EU-SOR, researchers have been restricted to imputing the regulation status of a worker
based on her reported occupation (e.g. Gittleman and Kleiner 2016; Koumenta et al. 2014).
While informative, the precision of such procedure is likely to be compromised by the
possibility that some of those classified as working in a licensed occupation were in fact not
licensed. For example, an engineer working in a multinational company in the automotive
industry might not need to be licensed, although the engineering profession is generally subject
to licensing. This highlights the difference between coverage and individual attainment of
licensed status. Moreover, the codes usually used to classify occupations do not necessarily
describe licensed professions, as defined by licensing regulations. By explicitly asking the
respondents to report their regulation status, EU-SOR enables us to address this measurement
3
problem. In this paper, we contribute to the literature in three ways. First, we provide the first
ever EU estimates of the prevalence of occupational licensing in the EU and estimate the wage
premium associated with licensing (the most restrictive form of regulation) and certification
(its less restrictive counterpart). Second, we decompose the wage premium using the Oaxaca-
Blinder decomposition and in doing so, provide more systematic descriptions of the wage gap
between licensed and non-licensed workers than currently offered. Third, we contribute to a
less well-developed line of enquiry in the occupational regulation literature, namely the impact
of licensing on the wage distribution. We do so using the Di Nardo, Fortin and Lemieux (1996)
decomposition method, which is itself a new approach to estimating such effects within the
occupational regulation literature. Throughout our analysis, we compare our findings with
those in the US, where the literature is more advanced, to establish any differences or
similarities between these two labour market contexts.
2. Related Literature
Entry requirements associated with licensing regulate the supply of labour in the market. This
is achieved in two ways. First, entry to the occupation is restricted to those practitioners whose
skills or character are above the minimum requirements. Second, regulators may revoke the
license if performance of incumbents is deemed to fall short of meeting the professional
standards. The implication of these arrangements is that the entry requirements associated with
licensing reduces the pool of practitioners thus creating monopoly rents within the occupation
(Pagliero 2013, 2011). As such, the wage effect of regulation is borne by the artificial creation
of entry barriers in the occupation, as opposed to the standard human capital returns in the
labour market. Research on the wage effect of regulation has a relatively long tradition, but
improvements in measurement have resulted in two empirical developments: (i) better
estimations of the wage premium associated with licensing (ii) detection of potential
differential effects of licensing on wage determination by occupation.
With regards to the former, in the US, Kleiner and Krueger (2013) using a self-reported
measure of regulation status find its effect on wages to be around the 11% mark. Gittleman and
Kleiner (2016) exploit the introduction of questions on occupational regulation in large-scale
national surveys. Using longitudinal data and a rich set of labour market controls they find
wage effects of about 7.5% considerably lower than previous estimates. Further, early studies
4
present licensing as having a homogeneous wage effect, without any consideration on the basis
of occupation and licensing regimes. More recently, researchers have become more alert to the
existence of some degree of heterogeneity in the effect of licensing on wages. According to
Kleiner (2013), the ability of occupations to capture rents depends on factors associated with
the political economy of the regulatory regime, such as the strictness of entry barriers or the
amount of time the occupation has been subject to regulation. Empirically, it is becoming
increasingly common to observe occupational variations on the wage effect of licensing. For
example, Timmons and Thornton use a cross-state U.S. survey radiologic technologists and
find the wage effect of licensing to be 6.9% (Timmons and Thornton 2007), while the same
figure for massage therapists stands at 16.2 % (Timmons and Thornton 2013).
In his comparison of average incomes across licensed and non-licensed occupations, Kleiner
(2000) calculates the licensing premium among dentists, and lawyers vis-à-vis other
comparable occupations. Despite many similarities in the educational licensing requirements
for dentists and lawyers, the wage effect is 30% for the former and 10% for the later. In the UK
context too, Koumenta et al. (2014) find significant variations in the wage premium amongst
dentists, pharmacists, accountants, architects, security guards, teachers and plumbers (ranging
from 9% to 19%). Although these studies allude to potential heterogeneity in the effect of
licensing on wages, they fall short of a systematic examination by occupational classification.
Finally, our data source enables us to go even further in understanding the effect of regulation
in the labour market by examining its effect on returns to education. Does licensing create a
distortion in the returns to education and if so, how does this pattern vary by the level of
educational attainment?
A less well-developed theme is the impact of licensing on different parts of the income
distribution. Licensing can result in the creation of rents through the monopoly effect discussed
above. Any such economic rents can aggravate income inequality if they are unequally
distributed amongst income groups, for example, those at the top of the income distribution
fare better than those at the bottom. Empirically, such a line of enquiry parallels that used to
study the effect of unionism (another labour market institution associated with rent-capture) on
wage dispersion. For example, in his classic work on the effects of unions, Freeman (1991)
finds that despite the inequality-increasing effect of unions on the difference between union
members and non-members, the overall effect of unions on income inequality is negative. Does
licensing have a similar effect on wage dispersion? Recent analyses by Kleiner and Krueger
5
(2013) and Gittleman and Kleiner (2016) for example find that licensing does not reduce wage
dispersion in the US labour market.
Also less common in empirical analysis is the effect of certification on wages. As a policy
alternative to licensing, certification is less restrictive in that it presents consumers with a
choice between practitioners whose credentials have been vetted by the state or a professional
body and those who do not meet such criteria. In theory, we would expect its effect on wages
to depend on the extent to which consumers are willing to pay a premium for practitioners
meeting these standards and is therefore difficult to determine a priori.
3. The survey and the sample
The data used for this study is based on the EU-SOR, the first ever survey commissioned by
the European Commission dedicated to capturing the extent of occupational regulation in the
EU. The questionnaire items are drawn from questions successfully tested in the US-based
Westat survey of regulated occupations (Kleiner and Krueger 2013), and further developed to
suit the specific research context. To test the final instrument, a pilot study was carried out
which suggested some revisions to the questionnaire, including shortening its length and some
minor adjustments to the item wording.
The survey covers the civilian population of the respective nationalities of the EU member
states, resident in each of the 28 member states, aged 15 years and over who at the time were
working or looking for a job. The survey was carried out in the period between March to April
2015 by means of telephone interviews (Computer Assisted Telephone Interviews) using a
multi-stage random probability sample design. In every country, respondents were called on
fixed lines and mobile phones and in each household, the respondent was drawn at random
following the ‘last birthday rule’.
A total of 26,640 individuals were interviewed providing data on their regulation status, the
characteristics of the regulation regime (e.g. entry and renewal requirements) as well as detailed
information on a variety of individual characteristics. The data has several important strengths
for our analysis. In addition to providing us with a self-reported measure of the regulation status
of the individual, the large sample size increases the external validity and precision of our
estimates. The breadth of information about individuals and their labour market circumstances
6
improves our ability to control for observable heterogeneity that might be correlated with
regulation status and earnings.
The response rate was 28%. The response rate is higher than the 17.9% obtained by Westat, its
American counterpart (Kleiner and Krueger 2013), analogous to that of commercial surveys
and in a context of declining response rates to telephone surveys (see for example Curtin et al.
2005) overall very satisfactory.4 We proceed to examine the credibility of our data by
comparing the mean proportion of respondents in the EU-SOR with those reported in the EU
Labour Force Survey5 (EULFS) for the same year (Table 1). In Column 3, we estimate the
difference in the means across a range of indicators relating to individual and labour market
characteristics. The two samples are closely matched, which increases our confidence
regarding the representativeness of our sample.
We repeat the same exercise in relation to earnings. Respondents in our survey are asked to
provide an estimate of their net monthly earnings from their main paid job (converted to euros
at the current exchange rate where necessary) and the number of hours of work in a typical
week. The number of hours is recorded as less than 15 hours per week, then in 5-hour intervals
up to 45, with the last category corresponding to more than 45 hours. We compute the hourly
net wage by dividing the reported wage by the estimated number of hours worked in a month
(4.35*the midpoint of each category of weekly hours worked). In Table 1, column 1 the mean
annual net wage is computed for a hypothetical full time worker working 40 hours per week
and 52 weeks per year. The corresponding figure from Eurostat (column 2) is computed starting
from gross wages and applying the average income tax for a single worker with average taxable
income. As before, the difference between the two calculations is negligible.
Table 1 about here
4 Nevertheless, it is important to consider the possibility of non-response errors. Using meta-analytic techniques of relevant studies, the literature finds no consistent relationship between response rates and non-response bias in large surveys (Groves 2006). Further, the higher response rates do not guarantee high data quality and should also be supplemented by measures of representativeness (Schouten et al. 2009). 5 Details about the EU Labour Force Survey can be found here: http://ec.europa.eu/eurostat/web/microdata/european-union-labour-force-survey
7
Finally, survey weights were developed by TNS (the organisation responsible for the
implementation of the survey) to compensate for variation in selection probabilities and non-
response bias.6
4. Methodology
We estimate cross sectional wage regressions of the general form
𝑌𝑖 = 𝑏0 + 𝑏1 𝐿𝑖𝑐𝑒𝑛𝑠𝑒𝑑𝑖 + 𝑋𝑖𝑏2 + 𝑢𝑖 (1)
where Yi denotes the log hourly wage of worker i, the matrix Xi includes gender, education,
union membership, work experience, working status, country, occupation, and industry fixed
effects. The coefficient b1 measures the impact of the indicator variable Licensed, which
measures whether worker i is subject to occupational licensing.
This approach assumes that the impact of occupational regulation is uniform across
occupations, and that regulation cannot affect the return to other individual characteristics (the
vector b2). This assumption is restrictive, since regulation may affect differently workers in
different occupations. Moreover, it may induce changes in the coefficients of other variables
such as education or work experience. A more general model allows for different coefficients
b1 and b2 for licensed and unlicensed workers.
Individuals in the survey can be partitioned into two exclusive groups denoted by g=L, N.
Individuals in group L are individuals who need a license to do their job, while those in group
N do not. The coefficients from the group-specific wage regressions
𝑌𝑔𝑖 = 𝛽𝑔0 + 𝑋𝑔𝑖𝛽𝑔1 + 𝑢𝑔𝑖 (2)
can be used to decompose the difference in average outcomes between group L and group N,
∆= 𝑌𝐿 − 𝑌𝑁 , (3)
6 For all countries surveyed, a national weighting procedure using marginal and intercellular weighting was carried out based on a comparison between the sample and the universe. The universe description was derived from Eurostat population data or from national statistics offices. In all countries, gender, age, region and size of locality were introduced in the iteration procedure. For international weighting (i.e. EU averages), the official population figures as provided by Eurostat or national statistic offices were used.
8
into the part explained by differences in characteristics X across the two groups (∆𝑋), and the
structural component (∆𝑆) that is due to differences in the coefficients 𝛽𝑔0 and 𝛽𝑔1 across the
two groups,
∆= ∆𝑋 + ∆𝑆 (4)
∆𝑋= (𝑋𝐿 − 𝑋𝑁)�̂�𝑁1
∆𝑆= (�̂�𝐿0 − �̂�𝑁0) + 𝑋𝐿(�̂�𝐿1 − �̂�𝑁1).
Hence, we can estimate by OLS the the wage regressions (2) and then decompose the overall
change in wages into what is driven by X, that is ∆𝑋, and what is driven by having a license,
∆𝑆 (Oaxaca 1973 and Blinder 1973). More generally, one can decompose the entire distribution
of wages and study the impact of licensing on any quantile of the wage distribution (DiNardo,
Fortin and Lemieux 1996).
Cross sectional wage regressions have the advantage of being representative of all occupations.
However, they are based on the usual assumption that E(u|X)=0. Hence, they cannot account
for unobserved characteristics that could be related to the selection into licensing occupations.
Decomposition methods rely on the same assumption. They provide interesting descriptive
results, which can be interpreted as causal only under this fairly strong assumption.
5. The prevalence of occupational licensing
The following two questions are used in the EU-SOR to classify workers into three groups: (i)
licensed; (ii) certified (or accredited), and (iii) unregulated:
7 In addition to specific training, interviewers received the following instructions: “A professional certification or
licence shows you are qualified to perform a specific job and may give you the right to enter a regulated profession
or professional association. Only include certifications or licences obtained by the respondent as an individual.
Examples include "licensed medical doctor" and "licensed taxi driver […]”.
“In addition to this education, do you have a professional certification, licence or did you
have to take an exam which is required to practice your occupation?”7
9
To distinguish between licensing and certification, those who answer 1 or 2 to the above
question were then asked:
A worker is classified as ‘licensed’ if she answers (1) or (2) in the first question and (2) in the
second. A worker is classified as ‘certified’ if she answers (1) or (2) in the first question and
(1) or (3) in the second, and ‘unregulated’ otherwise.
To ensure the validity of the survey instrument we derived the key questions from the
previously validated instrument used in the Westat survey in the US by Kleiner and Krueger
(2013). Second, the questionnaire was subject to piloting in all EU28 member states and no
issues were raised in relation to the questions on regulation status. Further, it is encouraging
that the ‘Don’t know/No answer’ responses are small, suggesting that respondents could
understand the question. Finally, the proportion of missing values in the data is similar to that
observed for other questions such as ‘economic activity of the firm or organization’ and
‘occupation’, again suggesting that the questions were well understood.
We estimate that just under half (43%) of the EU workforce is subject to some form of
regulation. In particular, 22% of workers are licensed (Table 2). The proportion of licensed
workers in the EU is broadly comparable to the most recent US estimates where, depending on
the data source, it ranges from 29% (Kleiner and Krueger 2013) to 20% (Gittleman et al.
8 Instructions to the interviewer: Refer to the respondent's specific occupation and personal circumstances. Refer
to the current laws and regulations affecting the respondent's occupation (current main paid job).
1. Yes…………(39.65%)
2. No – but currently in process of obtaining one…………. (0.95%)
3. No…………. (59.21%)
4. Don’t know/No answer ………(0.19%)
“Without this professional certification, licence or exam would you be legally allowed to
practice your occupation?”8
1. Yes…………(45.19%)
2. No………… (52.55%)
3. Don’t know/No answer ………(2.26%)
10
2018)9. It is also comparable to other dominant labour market institutions such as trade
unionism for which the latest estimates show an average level of union membership across the
EU (weighted by the numbers employed in the different member states) to be 23% (Fulton
2015). As a policy alternative to licensing, certification seems considerably more widespread
in the EU (22%) compared to Kleiner and Krueger’s (6%) and Gittleman et al.’s (2018)
estimates of 6% and 8% respectively. However, the former study only focuses on government
certification and the later only private certification, while our estimates include both
government and private certification. Hence, these figures are not directly comparable.
Table 2 about here
5.1. Who is licensed in the European Union?
We proceed with analysing the distribution of licensed, certified and unregulated workers based
on key demographic and employment characteristics (Table 3). We find that licensing is more
prevalent amongst those aged 40 to 64 years and male workers. Our results further indicate that
the incidence of licensing is higher amongst employees in the public sector (32%) and the self-
employed, and least common amongst those in private firms (16%). This is partly expected as
many occupations with high information asymmetries and potential to cause harm to others are
found in the public sector (e.g. medical occupations, teachers etc.) and amongst those in self-
employment (which is usually correlated with the provision of personal services e.g. plumbers,
lawyers etc.). This trend is further reflected in the distribution of licensing by industrial
classification, where we find that licensing is most prevalent in health and in public
administration and least common in the hotel, restaurants and retail industries. With respect to
occupational groups the incidence of licensing is highest amongst plant and machine operators
(35%), professionals (26%), and technicians and associated professionals (27%).
In the case of craft and related trades (20%) (but also to some extent for plant and machine
operators), we might be capturing the legacy of the guilds, the institution that for many
centuries provided the main route for entry to such occupations in Europe. Unsurprisingly, the
9 Kleiner and Krueger (2013) use direct measures of license attainment but their sample size and response rate are relatively small. Gittleman et al. (2018) use a large, nationally representative dataset, but the questions on credentials do not allow them to distinguish between licensing and certification with confidence.
11
lowest proportion of licensed workers is found amongst occupations where information
asymmetries and negative externalities are lowest (i.e. elementary occupations) and in
relatively ‘new’ occupational groups that lack a strong enough professional identity to organise
and lobby (i.e. managers). Finally, we find a relatively even spread of licensing and
certification among the various occupational groups.
Turning to education, licensing is common throughout the different levels of educational
attainment. It is common among individuals with lower secondary education (24.7%), post-
secondary education (26.8%), and advanced research qualification (24%). Interestingly, we
find no stark differences between licensing and certification on one hand and educational
attainment on the other. However, it is noteworthy that contrary to licensing and certification,
a large proportion of unregulated workers are concentrated in the low education category
(71%). Finally, licensing is slightly more common among union members than non-members,
most likely driven by the higher incidence of licensing in the public sector where unionisation
rates also tend to be high. Our results are broadly in line with the characteristics of licensed
workers observed by Kleiner and Krueger (2013) for the US who also find a higher proportion
of licensed workers in the public sector, amongst union members, and in the service industry10.
Table 3 about here.
To gauge the barriers to entry associated with working in a licensed occupation in the EU we
estimate the proportion of individuals that are subject to requirements relating to education, but
also additional hurdles such as entry examinations and work experience (Table 4). We find
passing an exam to be widespread, affecting around 86% of licensed individuals. While for the
majority work experience is not required (52%), over a quarter require work experience of
more than two years. With regards to education, either lower secondary, upper secondary or
university qualifications are required for about three quarters of those licensed, although the
proportion of those requiring no educational credentials is not insignificant (19%). Overall, we
find the restrictiveness of entry to licensed occupations in the EU to be comparable to that in
the US both in terms of the forms it takes but also the proportions of those affected. Of course
10 Due to differences in the classification of educational attainment, comparisons between the two are difficult.
12
the presence of barriers to entry in itself is neither a guarantee of high quality services on one
hand, nor a signal of occupational closure on the other since it is the level at which these
requirements are set that is likely to determine such outcomes. Nevertheless, it is suggestive of
the hurdles that individuals have to overcome to become licensed vis-à-vis those in unregulated
occupations.
Table 4 about here.
Our data further provides us with information about the distribution of licensed workers across
different member states (Table 5). Overall, the proportion of licensed workers ranges between
14% and 33% (Column 1). We find the largest proportion of licensed individuals in Germany
(33%), Croatia (31%), and Ireland (29%) and the smallest proportion in Sweden (15%), Latvia
(15%), and Denmark (14%). We further compare these data on individual attainment to the
number of regulated professions in the EU Database of Regulated Occupations (Column 3). 11
Interestingly, we do not find a statistically significant correlation between the proportion of
individuals subject to licensing and number of licensed occupations in each member state (r
=.13, p= n.s). For example, whilst the incidence of licensing within a member state might
initially appear low compared to others (as measured by the number of occupations subject to
licensing), the actual prevalence of licensing can be substantial if employment in these
categories is disproportionately large. We depict the geographical variability of the proportion
of licensed workers (Figure 1). Licensing is most prevalent in some Central and Eastern
European countries and there is no clear difference between North and South. Countries with
high proportion of licensing seem to be located along a diagonal from North-West to South-
East.
Table 5 about here
11 The database is compiled and maintained by the European Commission and it is accessible via this link: http://ec.europa.eu/growth/tools-databases/regprof/index.cfm?action=map_regulations
13
Figure 1 about here
6. The Wage effects of Occupational Licensing
To examine whether licensing is associated with higher pay, we present estimates of log wage
regressions (Table 6). In addition to the standard human capital controls, industry and detailed
occupational controls, country fixed effects are also included in the models. In our most basic
specification, the mean wage of licensed workers is about 9 log points higher than that of
unregulated workers. The coefficient of licensing then significantly drops as controls are
progressively included, suggesting that a large portion of the premium is coming from
educational endowments and other labour market characteristics, rather than licensing per se.
As such, in our more elaborate specifications, we find that having a licence is associated with
approximately 4% higher hourly wages (p<0.05). Such an effect is considerably lower in
magnitude than the 18% licensing wage premium found by Kleiner and Krueger (2013) and
closer to the more recent estimates of 10% by Kleiner and Vorotnikov (2018) and 7% by
Gittleman and Kleiner (2016).
Table 6 about here
6.2. Signalling or entry restrictions?
What is the mechanism that generates the observed wage gap between licensed and unlicensed
workers? Occupational licensing may affect wages by improving human capital and providing
a credible signal to consumers. However, it may also increase wages by restricting labour
supply. In our data, we can observe a set of workers who report to hold a certification (on top
of their educational achievements). This certification provides a signal of additional human
capital, without necessarily being associated with a restriction in labour supply, as it is not a
legal requirement. As a result of the restrictiveness of licensing, in a wage regression we expect
the coefficient of the indicator variable for licensed workers to be larger than that of certified
14
workers. The difference between the two provides a measure of the impact of the legally
enforced entry restrictions in increasing the wages of licensed workers.
The results from the wage regressions are shown in Table 7. We use the same specifications as
in Table 6, but add an indicator variable for certification status. Interestingly, as with licensing,
we find that certification has a positive and significant effect on wages. This is consistent with
certification being associated with higher skills and having some signalling value in the labour
market. However, in general, certification is associated with a smaller premium than licensing.
In column 6, this difference is about two percentage points (0.0308, p<0.05 for certification
and 0.0489, p<0.01 for licensing). Further, the difference between the coefficients of the
‘licensed’ and ‘certified’ dummies captures the impact of the legal requirement to hold a
license, which is the main difference between certification and licensing. In fact, both licensing
and certification provide a signal of worker quality. However, licensing also restricts entry into
specific labor markets to licensed workers and it is therefore associated with a higher wages.
We show a wage gap between licensed and unregulated workers of 4.89 log points, of which
37 percent can be attributed to entry restrictions and 63 percent to signalling12.
Table 7 about here
In the analysis that follows, we disaggregate the wage effects of licensing and certification by
occupation. This enables us to conduct within-occupation comparisons of those with licenses
and certificates on one hand, and those without at major occupation level (1-digit). The first
column in Table 8 includes controls for education, occupation, industry, work status, gender,
union indicators, age, experience, and experience squared. The second specification (column
2) includes the indicator variable for certification, while the third column includes interactions
of the certification dummy with an indicator variable for each occupation. Overall, we find that
licensing is having a differential effect by occupation. The effect is of highest magnitude for
craft and related trades occupations (19%) as well as elementary occupations (10%). Finally,
the difference between the coefficients of ‘licensed’ and ‘certified’ is very heterogeneous
12 We note that the standard error of the difference between the two coefficients is large.
15
across occupations. This difference is particularly large and statistically significant for craft
and related trades workers, elementary occupations, and service and sales workers.
Table 8 about here
6.3 Occupational licensing and the return to education
In line with the literature on returns to education (Becker 1993), we have shown that even after
controlling for human capital covariates in our wage regressions, the difference in wages due
to education remains large. Yet, an unexplored line of enquiry in the empirical literature is the
effect of occupational regulation on returns to education and in particular how this varies
between licensed, certified and unregulated workers. We address this by estimating wage
regressions similar to our previous models allowing for group-specific returns to education.
We split workers into three groups: elementary and lower secondary education, upper
secondary and tertiary (non-college) education, and college or higher. In Table 9, we report the
coefficients we obtain when we interact the ‘licensing’ and ‘certification’ indicator variables
with indicator variables for different levels of educational achievement. For unregulated
workers, the average wage grows with education, particularly for those with more than upper
secondary education. However, for licensed workers, we obtain different results (Figure 2).
While those with elementary education earn more than unregulated workers, after post-
secondary education we do not observe any growth in their returns to education. In fact, the
growth in the return to education for licensed workers with post-secondary education is zero.
From this, we conclude that there is no significant difference between the return to education
for a licensed and unlicensed worker with upper secondary or post-secondary education. This
pattern changes when we look at individuals with university degrees. Here, the return to a
university degree is much larger for licensed workers compared to their unregulated
counterparts, controlling for other observables. Hence, licensing flattens the returns to
education for low levels of education, while it increases the growth in the return to education
for university degrees and advanced research qualifications.
The differences in the returns to education between unregulated and regulated workers
diminish when we look at certification. In particular, certified workers experience a steady
growth in their wages as their education increases leading us to conclude that it is mainly
16
licensing that creates the distortion in the returns to education. This finding is consistent with
our previous results relating to the heterogeneous effects of licensing across different
occupational and educational categories.
Table 10 summarizes our results from the different wage regressions, with and without control
variables. The overall licensing wage gap (the difference between licensed and unregulated
workers) in columns 1 and 2 can be partly attributed to the signalling value of holding a license
(the certification gap, between 56 and 63 percent of the overall gap), the rest being attributable
to the legal requirement to hold a license (the legal requirement gap). Table 10 also reports the
same figures for different educational groups, highlighting the heterogeneity of the gap across
workers.
Table 9 about here
Figure 2 about here
Table 10 about here
6.3. Oaxaca-Blinder decomposition
The heterogeneity in wage premiums across occupations suggests that the linear model (1) may
not capture some of the heterogeneous effects of licensing regulations. In fact, our results so
far have highlighted the importance of heterogeneity of the licensing wage gap across
occupations and educational groups. We now turn to the Oaxaca-Blinder decomposition, which
does not constrain the effect of licensing to be constant for all workers, or even for workers
within the same occupation or educational group.
The estimated coefficients of model (2) on the two groups (licensed and non-licensed) are
reported in Table 11, together with the mean values of each variable. The table illustrates how
licensing distorts the relative wage of different occupational groups. For example, ‘Managers’
are the reference category, hence the value of the coefficient for this group is equal to zero by
assumption. Table 11 shows that licensed professionals earn on average 7 percent more than
licensed managers, licensed craft workers 3 percent less, and licensed workers in elementary
occupations 21 percent less.
Table 11 shows a very different pattern of relative wages for non-licensed workers. Non-
licensed professionals earn 5.8 percent less than non-licensed managers, non-licensed craft
workers earn 31.7 percent less, and non-licensed workers in elementary occupations 42.6
17
percent less. The occupations with the largest differences between the two figures are
‘Professionals’, ‘Craft and related trades’, ‘Service and sales workers’ and ‘Elementary
occupations’. These differences are consistent with the idea that licensing confers a wage
premium that is particularly large for some occupations.
Table 12 describes the results of the decomposition in equation (4) based on the estimated
coefficients. Table 12 shows that the overall difference in log wages between licensed and non-
licensed workers (about 9 percent) can be decomposed into the part that is due to characteristics
of the workers (composition effects) and that due to differences in regression coefficients (wage
structure effect). The wage structure effect can be interpreted as a generalized version of the
“wage premium” discussed in previous paragraphs.
The composition effect accounts for 61% of the overall difference. Differences in occupation,
age, and work experience are important determinants of the composition effect. The wage
structure effect, due to differences in the estimated coefficients, accounts for about 39% of the
overall effect. Differences in the coefficients of union membership, age, education dummies,
occupation dummies, and industry dummies are the most important contributors to the wage
structure effect. These results are in line with our previous results suggesting that the wage
premium associated with licensing is very different across occupations.
Table 11 about here
Table 12 about here
6. Occupational Licensing and Wage Dispersion
Our final analysis concerns the effect of licensing on the entire wage distribution. We do so
using the DiNardo, Fortin and Lemieux (1996) semiparametric decomposition method (DFL).
In Figure 2 we depict the distribution of log hourly wages in the EU for licensed and non-
licensed workers. We then decompose the difference between these two distributions into the
composition effect and the wage structure effect. In Figure 3 we show the distribution for
licensed workers and the estimated counterfactual density that would be obtained if these
workers had the same characteristics as their non-licensed counterparts. The composition effect
is equal to the difference between the two distributions. Finally, in Figure 4, we report the wage
18
distribution of non-licensed workers and the same counterfactual density. The difference
between these two distributions corresponds to the impact of licensing, holding constant the
characteristics of workers. This is the semiparametric version of the wage structure effect
introduced in equation (4).
Figure 3 about here
Figure 4 about here
Figure 5 about here
The results of the DFL decomposition can be used to compute statistics from the three
distributions. In Table 13, we report the standard deviation, the variance, and the distances
between selected quantiles. Differences in these statistics provide estimates of the composition
and wage structure effects. The wage structure effect of licensing implies a significant increase
in wage inequality as measured by the standard deviation of log hourly wages (0.02 log points).
Also the distance between the 99th and the 1st, the 95th and the 5th, the 90th and the 10th
percentiles are increased by licensing. Columns 6 and 7 show that the wage structure effect
leads to an increase in the dispersion of wages in both tails of the distribution. In line with
previous results on mean wages, we find that the median wage is increased by the wage
structure effect (column 8). Our results are in line with similar US studies by using different
sets of data and different methodologies (e.g. Kleiner and Krueger (2013); Gittleman and
Kleiner (2016); Kleiner and Vorotnikov (2018).
Table 13 about here
7. Conclusions
19
We present the first ever estimates of the prevalence and wage effects of occupational
regulation in the EU. We find that licensing affects about 22 percent of workers in the EU,
albeit with significant variability across member states and occupations, leading us to conclude
that it is an important labour market institution in this context. We further show that it has an
effect on wages. In particular, licensing is associated with an aggregate wage premium of about
4% after accounting for observable characteristics of the workers, country and occupation fixed
effects. This figure is somewhat lower than current US estimates. Overall, our analysis
attributes approximately one third of the licensing wage gap to entry restrictions and the
remaining to signalling.
We further present the first systematic analysis of its heterogeneous effect by occupation- for
some groups such as crafts for example it can be as high as 19% while for professionals it drops
to 6%- and by level of educational attainment. With regards to the latter, licensing compresses
the returns to education for low educated individuals but increases the growth in the return to
education for those with university degrees and above. Moreover, licensing disproportionately
benefits those at the top of the income distribution, increasing the dispersion of wages. As such,
occupational licensing differs from unionization, which is known to reduce wage dispersion.
To the extent that pay inequality in the labour market is a concern for policy makers, then
licensing requires further consideration.
We also account for the possible effect of certifications that are not legally required to practice
an occupation but may signal the existence of labour market skills not fully captured by
conventional education. Our estimates are in line with the most recent findings in the US in
that certification is also associated with a wage premium, but of comparable magnitude to that
of licensing. In that sense, certification is perhaps a better policy alternative in that it improves
the skills of practitioners, signals the existence of a minimum standard to consumers while
allowing them to choose whether they are willing to pay the premium associated with using a
certified practitioner.
A key novelty of our approach is our ability to estimate the incidence of regulation based on
self-reported measures of regulation status rather than rely on inferences from occupational
classifications. However, the nature of our data does not allow us to rule out the possibility of
selection bias on unobservables. Nevertheless, the limitations of the current study open up
fruitful avenues for future research, particularly in the form of devising ways to explore
exogenous variation in the licensing status.
20
References
Becker, Gary (1993), Human Capital: A Theoretical and Empirical Analysis, with Special
Reference to Education, 3rd edition, Chicago: The University of Chicago Press.
Blinder, Alan S. (1973) Wage Discrimination: Reduced Form and Structural Estimates, The
Journal of Human Resources, Vol. 8, No. 4, pp. 436-455.
Curtin et al (2005) Changes in Telephone Survey nonresponse over the past quarter century,
Public Opinion Quarterly, 69
DiNardo, John, Nicole M. Fortin, and Thomas Lemieux. 1996. “Labor Market Institutions and
the Distribution of Wages, 1973-1992: A Semiparametric
Approach”. Econometrica 64 (5): 1001–44.
Forth, J., Bryson, A., Humphris, A., Kleiner, M. and Koumenta, M. (2011) A Review of
Occupational Regulation and its Impact, UK Commission for Employment and
Skills, London
Fulton, L. (2015) Worker representation in Europe, available at https://www.worker-
participation.eu/National-Industrial-Relations/Across-Europe/Trade-Unions2
Freeman, R. B. (1991). How much has de-unionisation contributed to the rise in male earnings
inequality? (No. w3826). National Bureau of Economic Research.
Friedman, M. (1962). Capitalism and freedom: With the assistance of Rose D. Friedman.
University of Chicago Press.
Graddy, E. (1991). Toward a general theory of occupational regulation. Social science
quarterly, 72(4), 676-695.
Groves 2006 Nonresponse rates and nonresponse bias in household surveys, public opinion
quarterly 70
Gittleman, M., & Kleiner, M. M. (2016). Wage effects of unionization and occupational
licensing coverage in the United States. ILR Review, 69(1), 142-172.
21
Kleiner, M. M. (2000). Occupational licensing. The Journal of Economic Perspectives, 14(4),
189-202.
Kleiner, M. and Krueger A. (2013) ‘Analysing the Extent and Influence of Occupational
Licensing on the Labor Market, Journal of Labor Economics, 31, 2.
Kleiner, M.M. and Vorotnikov, E. (2018) Analyzing Occupational Licensing Among the
States, Journal of Regulatory Economics, forthcoming
Koumenta, M., Humphris, A., Kleiner, M. and Pagliero, M. (2014) Occupational Regulation
in the UK and EU: Prevalence and Labour Market Impact, Department for Business,
Innovation and Skills, London
Maurizi, A. (1974). Occupational licensing and the public interest. Journal of Political
Economy, 82(2, Part 1), 399-413.
Oaxaca, Ronald (1973), Male-Female Wage Differentials in Urban Labor Markets,
International Economic Review, Vol. 14, No. 3, pp. 693-709
Pagliero, M. (2013) ‘The Impact of Potential Labour Supply on Licensing Exam Difficulty’,
Labour Economics, 25, 141-152.
Pagliero, M. (2011). What is the objective of professional licensing? Evidence from the US
market for lawyers. International journal of Industrial organization, 29(4), 473-
483.
Schouten et al (2009) Indicators for the representativeness of survey response, Survey
Methodology 35(1)
Shapiro, C. (1986) ‘Investment, Moral Hazard and Occupational Licensing’, Review of
Economic Studies, 53, 843-62.
Timmons, Edward and Robert Thornton. "The Licensing of Barbers in the US," British
Journal of Industrial Relations. 48(4), 2010.
22
TABLES
Table 1. Comparison of characteristics of EU workers in the Survey of Occupational Licensing and in the Eurostat Labour Force Survey (civilian
employment)
Mean proportion of workers EU-SOR Data Eurostat data difference
by educational attainment:
Less than primary, primary and lower secondary education (levels 0-2)
0.169 0.183 -0.014
Upper secondary and post-secondary non-tertiary education (levels 3 and 4)
0.439 0.481 -0.042
Tertiary education (levels 5-8) 0.392 0.332 0.060
0.000
by gender Male 0.542 0.540 0.002
0.000
by age From 15 to 39 years 0.436 0.427 0.009
From 40 to 64 years 0.543 0.550 -0.007
65 years or over 0.021 0.023 -0.003
0.000
by type of employment Employees 0.851 0.850 0.001
Self-employed persons with employees (employers)
0.043 0.042 0.001
23
Self-employed persons without employees (own-account workers)
0.105 0.108 -0.002
0.000
by industry Agriculture 0.037 0.046 -0.009
Manufacturing and construction 0.225 0.240 -0.015
Education and health 0.225 0.185 0.040
Trade 0.211 0.267 -0.056
Finance and professional services and other services
0.225 0.194 0.031
Public administration 0.078 0.069 0.009
by occupation Managers 0.100 0.060 0.040
Professionals 0.276 0.188 0.088
Technicians and associate professionals
0.158 0.158 0.000
Clerical support workers 0.096 0.097 -0.001
Service and sales workers 0.140 0.167 -0.027
Skilled agricultural, forestry and fishery
0.016 0.039 -0.022
Crafts and related workers 0.107 0.116 -0.009
Plant and machine operators and assemblers
0.051 0.074 -0.022
Elementary occupations 0.055 0.092 -0.037
Armed forces occupations dropped 0.006 n/a
No response dropped 0.002 n/a
Mean Net Earnings 22,037 23,142 -1,105
24
Table 2. Proportion of licensed and certified workers in the European Union.
Proportion Std.
Error Licensed 0.219 0.0048 Certified 0.217 0.0049
Unregulated 0.564 0.0058 Source: EU Survey of Regulated Occupations. Civilian employed population age 15 or older.
25
Table 3. Proportion of licensed, certified, and unregulated workers by worker
characteristics.
Licensed Certified Unregulated
Gender Male 0.218 0.202 0.551
Female 0.220 0.230 0.580 Age From 15 to 39 years 0.195 0.205 0.600
From 40 to 64 years 0.239 0.224 0.537
65 years or over 0.200 0.286 0.514
Employment status Employee in a private firm 0.159 0.217 0.624 Employee in public/non-profit sector 0.322 0.217 0.462
Self-employed with employees 0.241 0.213 0.545
Self-employed without employees 0.220 0.217 0.563
Occupation Managers 0.127 0.210 0.663
Professionals 0.263 0.229 0.508 Technicians and associate professionals 0.271 0.236 0.493
Clerical support workers 0.145 0.169 0.686
Service and sales workers 0.218 0.214 0.568 Skilled agricultural, forestry and fishing 0.160 0.239 0.602
Craft and related trades workers 0.199 0.268 0.533 Plant and machine operators and assemblers 0.349 0.176 0.475
Elementary occupations 0.105 0.153 0.742
Industry
Agriculture (A) 0.141 0.211 0.649
Manufacturing of products (B, C) 0.128 0.228 0.644
Construction or energy (D, E, F) 0.213 0.268 0.519
Wholesale or retail trade 0.106 0.199 0.695
Hotels and restaurants (I) 0.096 0.206 0.698
Transportation and communication 0.267 0.212 0.520
Finance, real estate (K, L) 0.242 0.217 0.541
Public administration (O) 0.349 0.198 0.453
Education (P) 0.271 0.226 0.503
Health and social work (Q) 0.367 0.196 0.438
26
Professional services 0.209 0.219 0.571
Cultural activities 0.148 0.210 0.641 Education Primary 0.123 0.165 0.712
Lower secondary 0.247 0.286 0.467
Upper secondary 0.209 0.208 0.583
Post-secondary 0.268 0.211 0.521 University (undergraduate/postgraduate) 0.215 0.204 0.580
PhD/advanced search qualification 0.240 0.163 0.597
Trade union membership Yes 0.280 0.224 0.496
No 0.201 0.214 0.585
Source: EU Survey of Regulated Occupations. Civilian employed population age 15 or older.
27
Table 4. Requirements to obtain a license.
Proportion of Licensed Workers
subject to the specific requirement Std. error
Specific requirement:
Entry examination 0.864 0.009
Educational requirement:
No requirement 0.191 0.010
Primary education 0.024 0.003
Lower secondary 0.209 0.011
Upper secondary 0.279 0.011
Post-secondary education 0.046 0.006
University 0.247 0.011
PHD/ adv. research 0.005 0.001
Experience requirement:
No work experience is/was required 0.524 0.013
Up to a year 0.094 0.006
More than 1 year up to 2 years 0.094 0.007
Longer than 2 years 0.288 0.012 Source: EU Survey of Regulated Occupations. Civilian employed population age 15 or older.
28
Table 5. The proportion of licensed workers by country.
Proportion of Licensed Workers
a (1)
Std. error (2)
Proportion of Licensed
Occupationsb (3)
Germany 0.329 0.017 67
Croatia 0.312 0.018 210
Ireland 0.293 0.019 63
Slovakia 0.268 0.016 308
Hungary 0.262 0.016 353
Netherlands 0.246 0.016 80
Czech Republic 0.244 0.015 351
Austria 0.222 0.015 144
Greece 0.218 0.016 n/a
Romania 0.217 0.018 132
Bulgaria 0.213 0.017 98
Luxemburg 0.210 0.020 21
Poland 0.205 0.014 345
Slovenia 0.202 0.015 125
United Kingdom 0.195 0.015 103
Italy 0.193 0.015 165
Estonia 0.192 0.014 98
Cyprus 0.185 0.021 96
Lithuania 0.175 0.013 70
Malta 0.172 0.019 n/a
Finland 0.167 0.013 127
Belgium 0.166 0.014 111
Spain 0.166 0.015 107
Portugal 0.165 0.014 202
France 0.160 0.013 238
Sweden 0.153 0.013 21
Latvia 0.151 0.012 266
Denmark 0.140 0.012 149 a Source: EU Survey of Regulated Occupations. Civilian employed population age 15 or older.
bSource: EU Database of Regulated Occupations.
29
Table 6. Coefficients from log wage regressions.
(1) (2) (3) (4) (5) (6)
licensed 0.0919*** 0.0509*** 0.0438*** 0.0397*** 0.0347** 0.0382**
(0.0224) (0.0151) (0.0148) (0.0147) (0.0149) (0.0151)
union 0.0385** 0.00591 0.0151 0.0125 0.0121
(0.0159) (0.0157) (0.0161) (0.0157) (0.0156)
male 0.174*** 0.173*** 0.165*** 0.148*** 0.149***
(0.0129) (0.0129) (0.0135) (0.0135) (0.0139)
age 0.0105*** 0.00669*** 0.00637*** 0.00628*** 0.00633***
(0.000682) (0.000793) (0.000761) (0.000735) (0.000725)
Lower secondary education (usually age 11-15) 0.142*** 0.133*** 0.0931** 0.0845* 0.0869**
(0.0461) (0.0466) (0.0432) (0.0437) (0.0442)
Upper secondary education (usually age 16-19) 0.249*** 0.235*** 0.161*** 0.151*** 0.155***
(0.0423) (0.0429) (0.0405) (0.0409) (0.0413)
Post-secondary education (not university) 0.314*** 0.298*** 0.183*** 0.174*** 0.174***
(0.0489) (0.0490) (0.0470) (0.0473) (0.0481)
University (undergraduate and post-graduate) 0.579*** 0.570*** 0.378*** 0.359*** 0.355***
(0.0428) (0.0434) (0.0425) (0.0429) (0.0432)
PHD/ advanced research qualification 0.800*** 0.811*** 0.583*** 0.577*** 0.571***
(0.0536) (0.0550) (0.0544) (0.0540) (0.0542)
Experience 0.0180*** 0.0151*** 0.0146*** 0.0141***
(0.00240) (0.00233) (0.00231) (0.00225)
Experience2/1,000 -0.314*** -0.271*** -0.260*** -0.254***
(0.0730) (0.0712) (0.0703) (0.0683)
Employee in public sector or non-profit -0.000385 0.0253 0.0256
(0.0137) (0.0188) (0.0185)
30
Self-employed with employees 0.182*** 0.194*** 0.202***
(0.0527) (0.0527) (0.0528)
Self-employed without employees -0.0653* -0.0464 -0.0435
(0.0393) (0.0385) (0.0362)
Country f.e? yes yes yes yes yes
Occupation controls? yes yes Industry controls? yes yes
Detailed occupation controls? yes
Observations 16,067 16,027 15,952 15,952 15,952 15,796
R-Squared 0.002 0.668 0.675 0.699 0.706 0.710
Note: The dependent variable is the log of hourly wage. Omitted indicator variables: Primary education, Employee in private firm or business.
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.
31
Table 7. Coefficients from log wage regressions (licensing and certification).
(1) (2) (3) (4) (5) (6)
licensed 0.109*** 0.0639*** 0.0571*** 0.0504*** 0.0452*** 0.0489***
(0.0232) (0.0157) (0.0155) (0.0154) (0.0155) (0.0157)
certified 0.0612** 0.0421** 0.0434** 0.0327** 0.0316** 0.0308*
(0.0252) (0.0172) (0.0171) (0.0163) (0.0159) (0.0158)
union 0.0368** 0.00415 0.0143 0.0116 0.0112
(0.0158) (0.0157) (0.0160) (0.0156) (0.0156)
male 0.173*** 0.172*** 0.164*** 0.148*** 0.148***
(0.0129) (0.0129) (0.0136) (0.0135) (0.0139)
age 0.0104*** 0.00660*** 0.00630*** 0.00622*** 0.00628***
(0.000674) (0.000784) (0.000755) (0.000729) (0.000719)
Lower secondary education (usually age 11-15) 0.140*** 0.131*** 0.0913** 0.0829* 0.0856*
(0.0461) (0.0465) (0.0431) (0.0437) (0.0442)
Upper secondary education (usually age 16-19) 0.248*** 0.234*** 0.160*** 0.151*** 0.155***
(0.0424) (0.0429) (0.0405) (0.0408) (0.0413)
Post-secondary education (not university) 0.313*** 0.297*** 0.183*** 0.174*** 0.174***
(0.0489) (0.0490) (0.0470) (0.0472) (0.0481)
University (undergraduate and post-graduate) 0.578*** 0.569*** 0.379*** 0.360*** 0.356***
(0.0428) (0.0434) (0.0425) (0.0429) (0.0433)
PHD/ advanced research qualification 0.803*** 0.813*** 0.587*** 0.581*** 0.575***
(0.0538) (0.0552) (0.0545) (0.0542) (0.0544)
Experience 0.0180*** 0.0151*** 0.0146*** 0.0141***
(0.00241) (0.00234) (0.00232) (0.00225)
Experience2/1,000 -0.314*** -0.271*** -0.260*** -0.254***
(0.0733) (0.0714) (0.0705) (0.0685)
32
Employee in public sector or non-profit -0.00228 0.0242 0.0246
(0.0138) (0.0190) (0.0186)
Self-employed with employees 0.182*** 0.194*** 0.202***
(0.0527) (0.0527) (0.0528)
Self-employed without employees -0.0661* -0.0468 -0.0436
-0.0393 (0.0385) (0.0363)
Country f.e? yes yes yes yes yes
Occupation controls? yes yes Industry controls? yes yes
Detailed occupation controls? yes
Observations 16156 16116 16041 16041 16041 15875
R-squared 0.003 0.668 0.676 0.699 0.706 0.711
Licensed - Certified 0.0473825 0.021733 0.013736 0.017663 0.013588 0.018093
s.e. (0.029455) (0.0199376) 0.01948 0.018712 0.018886 0.018885 Note: The dependent variable is the log of hourly wage. Omitted indicator variables: Primary education, Employee in private firm or business. Robust standard errors in parentheses ***
p<0.01, ** p<0.05, * p<0.1.
33
Table 8. Coefficients from log wage regressions (licensing and certification) with
interactions.
(1) (2) (3)
licensed x managers -0.0933 -0.08471 -0.0781
(0.0677) (0.0679) (0.0693)
licensed x professionals 0.046553 0.057031 0.062564
(0.0255) (0.0257) (0.0262) licensed x Technicians and associate professionals -0.0125 0.000237 -0.00051
(0.0317) (0.0325) (0.0351)
licensed x Clerical support workers -0.00907 -0.00032 0.000717
(0.0651) (0.0649) (0.0653)
licensed x Service and sales workers 0.078359 0.089778 0.086503
(0.0301) (0.0305) (0.0322)
licensed x Skilled agricultural 0.030306 0.041184 0.037242
(0.2351) (0.2360) (0.2323)
licensed x Craft and related trades workers 0.172144 0.184986 0.192386
(0.0396) (0.0400) (0.0419)
licensed x Plant and machine operators -0.04952 -0.03835 -0.05984
(0.0535) (0.0537) (0.0559)
licensed x Elementary occupations 0.120484 0.126519 0.098676
(0.0809) (0.0808) (0.0822)
certified 0.032714
(0.0160) certified x managers 0.0645
(0.0553)
certified x professionals 0.050384
(0.0354)
certified x Technicians and associate professionals 0.032273
(0.0346)
certified x Clerical support workers 0.041332
(0.0461)
certified x Service and sales workers 0.025043
(0.0450)
certified x Skilled agricultural 0.021177
(0.2102)
certified x Craft and related trades workers 0.056795
(0.0377)
certified x Plant and machine operators -0.03404
(0.0434)
certified x Elementary occupations -0.12061
(0.0710)
Difference between licensing and certification:
34
Managers -0.11742 -0.14267
(0.0688) (0.0794)
Professionals 0.024317 0.01218
(0.0285) (0.0379)
Technicians and associate professionals -0.03248 -0.03278
(0.0330) (0.0366)
Clerical support workers -0.03304 -0.04062
(0.0667) (0.0760)
Service and sales workers 0.057064 0.06146
(0.0320) (0.0451)
Skilled agricultural 0.00847 0.016066
(0.2365) (0.3004)
Craft and related trades workers 0.152272 0.135591
(0.0410) (0.0462)
Plant and machine operators -0.07106 -0.0258
(0.0548) (0.0597)
Elementary occupations 0.093805 0.219284
(0.0822) (0.0979) Note: The dependent variable is the log of hourly wage. The coefficients of education, occupation, industry, work status,
gender, union indicators, age, experience, and experience squared are not reported. Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
35
Table 9. Regression coefficients with interactions with education dummies.
(1)
licensed 0.131***
(0.0376)
certified 0.0776**
(0.0367) Licensed x Upper secondary education and Post-secondary education (not university) -0.126***
(0.0427)
Licensed x University education (and higher) -0.0708
(0.0436) Certified x Upper secondary education and Post-secondary education (not university) -0.0684
(0.0430)
Certified x University education (and higher) -0.0493
(0.0476)
Upper secondary education and Post-secondary education (not university) 0.131***
(0.0254)
University education (and higher) 0.320***
(0.0283)
union 0.0104
(0.0157)
male 0.150***
(0.0140)
age 0.00639***
(0.000716)
Experience 0.0140***
(0.00225)
Experience2/1,000 -0.255***
(0.0684)
Employee in public sector or non-profit 0.0238
(0.0186)
Self-employed with employees 0.204***
(0.0530)
Self-employed without employees -0.0426
(0.0366)
Country f.e? yes
Occupation controls? Industry controls? yes
Detailed occupation controls? yes
Observations 15,795
R-squared 0.709 Note: The dependent variable is the log of hourly wage. See Figure 1 for graphical representation. Omitted indicator variables: Primary and lower secondary education, Employee in private firm or business. Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
36
Table 10. The licensing, certification, and legal requirement wage gaps (summary table).
(1) (2) (3)
Overall licensing wage gap (Licensed - Unregulated) 0.109 0.0489
Primary and lower secondary education 0.131
Upper secondary and Post-secondary education (not university)
0.005
University education (and higher) 0.060
Certification gap (Certified - Unregulated) 0.0612 0.0308
Primary and lower secondary education 0.078
Upper secondary and Post-secondary education (not university)
0.009
University education (and higher) 0.028
Legal requirement gap (Licensed - Certification) 0.0478 0.0181
Primary and lower secondary education 0.053
Upper secondary and Post-secondary education (not university)
-0.004
University education (and higher) 0.032
no controls
controls controls
Note: the table summarizes estimated wage differences across groups of workers (see Tables 7-10) without controlling for
worker characteristics (column 1) and controlling for worker characteristics (columns 2 and 3). In column 3, wage
differences are reported by educational group.
37
Table 11. Means and regression coefficients from log wage regressions.
(1) (2) (3) (4) (5) (6)
Licensed Non-
licensed Licensed Licensed Non-
licensed Non-licensed
Means Means coeff. s.e. coeff. s.e.
union 0.317 0.237 -0.032 0.026 0.026 0.019
age 43.291 41.567 0.004 0.001 0.007 0.001
Experience 12.560 10.863 0.015 0.003 0.015 0.003
Experience2/1,000 0.271 0.225 -0.167 0.082 -0.294 0.088
male 0.559 0.535 0.123 0.026 0.152 0.015
Education (primary education omitted):
Lower secondary education (usually age 11-15) 0.153 0.139 -0.036 0.106 0.082 0.047
Upper secondary education (usually age 16-19) 0.353 0.366 0.004 0.099 0.163 0.044
Post-secondary education (not university) 0.075 0.064 0.002 0.118 0.197 0.050
University (undergraduate and post-graduate) 0.381 0.378 0.264 0.104 0.351 0.046
PHD/ advanced research qualification 0.023 0.022 0.455 0.119 0.579 0.060
Occupation (managers omitted):
Professionals 0.320 0.260 0.068 0.063 -0.058 0.028
Technicians and associate professionals 0.188 0.145 -0.032 0.067 -0.141 0.030
Clerical support workers 0.064 0.109 -0.148 0.085 -0.235 0.030
Service and sales workers 0.150 0.141 -0.144 0.065 -0.338 0.031
Skilled agricultural 0.009 0.014 -0.353 0.221 -0.383 0.085
Craft and related trades workers 0.098 0.111 -0.031 0.074 -0.317 0.032
Plant and machine operators 0.085 0.046 -0.192 0.091 -0.275 0.033
Elementary occupations 0.027 0.062 -0.209 0.095 -0.426 0.041
Industry (agriculture omitted):
Manufacturing of products 0.078 0.148 0.164 0.082 0.277 0.061
38
Construction or energy 0.083 0.085 0.207 0.087 0.304 0.064
Wholesale or retail trade 0.051 0.137 0.068 0.089 0.193 0.062
Hotels and restaurants 0.015 0.037 -0.055 0.113 0.081 0.076
Transportation and communication 0.068 0.056 0.215 0.089 0.326 0.064
Finance, real estate 0.044 0.040 0.104 0.110 0.336 0.067
Public administration 0.144 0.073 0.148 0.084 0.199 0.062
Education 0.137 0.103 0.079 0.088 0.163 0.067
Health and social work 0.220 0.108 0.061 0.083 0.208 0.063
Professional services (e.g. legal) 0.120 0.141 0.082 0.085 0.268 0.062
Cultural activities (including sport) 0.020 0.041 -0.040 0.106 0.092 0.067
Work status (employee in private firm omitted):
Employee in public sector or non-profit 0.496 0.294 0.022 0.037 0.029 0.022
Self-employed with employees 0.036 0.033 0.195 0.089 0.205 0.062
Self-employed without employees 0.084 0.081 0.001 0.061 -0.053 0.046
constant 1.000 1.000 1.967 0.158 1.674 0.088 Note: The table reports the mean of the variables for licensed and non-licensed workers in columns 1 and 2. Columns 3 and 4 report the coefficients and standard errors of a wage regression
for licensed workers. Columns 5 and 6 report the coefficients and standard errors of a wage regression for non-licensed workers. Country fixed effects are not reported. Number of obs. =
15951.
Use coefficients to produce the
39
Table 12. Oaxaca-Blinder decomposition results.
(1) (2) (3) (4)
Estimated values s.e. t Proportion of total
Differential
Predicted log wage licensed 2.1669 0.0195 111.36 Predicted log wage non-licensed 2.0773 0.0108 192.09
Difference 0.0896 0.0223 4.03 100%
Explained composition effects attributable to
Union 0.0021 0.0016 1.33 Age 0.0119 0.0032 3.69 Work experience 0.0113 0.0026 4.33 Gender 0.0037 0.0025 1.47 Education 0.0030 0.0041 0.72 Occupation 0.0085 0.0048 1.78 Industry 0.0027 0.0056 0.47 Work status 0.0063 0.0045 1.39 Country 0.0050 0.0157 0.32
Total 0.0545 0.0183 2.98 61%
Unexplained wage structure effect attributable to
Union -0.0184 0.0102 -1.81 Age -0.1367 0.0671 -2.04 Work experience 0.0320 0.0276 1.16 Gender -0.0159 0.0168 -0.95 Education -0.1250 0.1081 -1.16 Occupation 0.1365 0.0620 2.2 Industry -0.1195 0.0953 -1.25 Work status 0.0011 0.0236 0.05 Country -0.0123 0.0377 -0.33 Constant 0.2932 0.1805 1.62
Total 0.0351 0.0154 2.28 39%
Note: The table uses non-licensed workers as reference group. The estimated coefficients and mean values of the variables are reported in Table xx. Robust standard errors are reported in column 2.
l40
Table 13. Licensing and wage inequality: aggregate decomposition results.
(1) (2) (3) (4) (5) (6) (7) (8) sd var p99-p1 p95-p5 p90-p10 p50-p5 p95-p50 Median
Licensed workers 0.8569 0.7343 3.7768 2.8273 2.2264 1.4093 1.418 1.9841
Non-licensed workers 0.8451 0.7142 3.6706 2.7049 2.2114 1.3481 1.3568 1.8417
Counter factual 0.8698 0.7565 3.8243 2.8685 2.2476 1.4118 1.4568 1.8679
Total effect 0.0118 0.0201 0.1062 0.1224 0.015 0.0612 0.0612 0.1424
Composition effect -0.0128 -0.0221 -0.0475 -0.0412 -0.0212 -0.0025 -0.0387 0.1162
Wage structure effect 0.0246 0.0422 0.1537 0.1637 0.0362 0.0637 0.0999 0.0262
Note: DFL decomposition results. The explanatory variables include union, education, occupation, industry, work status, country indicators, age, experience, experience squared.
l41
FIGURES
Figure 1. The proportion of licensed workers in the EU.
l42
Figure 2. The wage effects of licensing and certification by education.
Note: Figure reports the wage effects of licensing and certification relative to unregulated workers for workers with
elementary or lower secondary education, upper secondary and tertiary (non-college) education, and college or higher.
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
95 perc
5perc
coeff
l43
Figure 3. Log wage distribution for licensed and non-licensed workers.
Figure 4. The composition effect on the log wage distribution.
l44
Figure 5. The wage structure effect on the log wage distribution.